Inference of the infection status of individuals using longitudinal testing data from cryptic populations: Towards a probabilistic approach to diagnosis

Svetlana N Buzdugan,Timothée Vergne,Richard J Delahay,Julian A Drewe,Vladimir Grosbois

doi:10.1038/s41598-017-00806-4

Abstract

Effective control of many diseases requires the accurate detection of infected individuals. Confidently ascertaining whether an individual is infected can be challenging when diagnostic tests are imperfect and when some individuals go for long periods of time without being observed or sampled. Here, we use a multi-event capture-recapture approach to model imperfect observations of true epidemiological states. We describe a method for interpreting potentially disparate results from individuals sampled multiple times over an extended period, using empirical data from a wild badger population naturally infected with Mycobacterium bovis as an example. We examine the effect of sex, capture history and current and historical diagnostic test results on the probability of being truly infected, given any combination of diagnostic test results. In doing so, we move diagnosis away from the traditional binary classification of apparently infected versus uninfected to a probability-based interpretation which is updated each time an individual is re-sampled. Our findings identified temporal variation in infection status and suggest that capture probability is influenced by year, season and infection status. This novel approach to combining ecological and epidemiological data may aid disease management decision-making by providing a framework for the integration of multiple diagnostic test data with other information.

Highlights

Detecting infection with confidence in individuals can be difficult when diagnostic tests are imperfect because the true state of infection is not, or is only partially, determinable[1]
In this study we used a unique long-term empirical dataset of badger capture and testing histories to estimate the probability of infection with M. bovis, for each individual badger, given any combination of results from three imperfect diagnostic tests, and accounting for each animal’s capture and test result history
For long-term surveillance of disease in this population, applying a probabilistic approach to infection status represents a clear improvement on current practice where a positive or negative diagnosis is usually based on imperfect tests applied at a single sampling point[26, 27]

Summary

Introduction

Detecting infection with confidence in individuals can be difficult when diagnostic tests are imperfect because the true state of infection is not, or is only partially, determinable[1]. The problem of imperfect state observation (which includes false positive and false negative diagnostic test results) was discussed by Conn and Cooch[10] They used a multi-event CRC modelling framework to incorporate data from uncertain or unknown states, a process they claimed increased the precision of disease dynamic parameter estimates[10]. Such a method may help deal with uncertainty in ascertaining state, it doesn’t deal with misclassification bias because of the implicit assumption that the state of an individual (e.g. its infection status) is determined accurately if information on state is obtained at all[10]. This wastes potentially useful information for inferring current infection status in individuals that are repeatedly sampled over time (for example, those with a chronic or recurring infection), given the imperfect performance of diagnostic tests

Objectives

Methods

Results

Conclusion